Coordinator: Vincent Labatut (LIA)

WP2 covers the work related to the extraction and analysis of the collaboration networks.
First, Task 2.1 consists in exploring the different ways these graphs can be built based on the database
elaborated during WP1, as well as conducting their descriptive analysis. Obviously, this requires a
stable version of the database, which is why it must start after making sufficient progress on Task 1.3 (Database Creation).
Nevertheless, it is still possible to start working on Task 2.1 even if the database is not completely
achieved, e.g. by focusing on a part of the data. It is worth stressing that the two tasks are meant to
feed each other: the descriptive analysis of the collaboration networks could uncover some
observations likely to affect the constitution of the database. For instance, the networks could help
identify zones of the data space that are insufficiently covered by the database, and require some
additional data collection work.
The final goal of this WP is to identify features of the collaboration networks that differ
depending on gender, and to cross-reference them with our other variables of interest: research
domain and country. This can be formulated as a prediction problem, in which one wants to train a
model into determining the gender of a researcher. We will first focus on leveraging interactions and
individual information to perform this prediction in Task 2.2, before turning to the representation of
time to better capture the system dynamics, in Task 2.3. The former task includes a significant survey
of the literature, and can therefore start even before the extraction of the networks (Task 2.1), which
gives us some flexibility.
Chronologically, WP2 will be located between WP1 and WP3, as it will require the data
produced during WP1, and will produce the statistics and observations that will feed the work in WP3.
All its tasks will be addressed by LIA and ERIC, as they require mostly Computer Science expertise. The
doctoral candidate attributed to WP2 and jointly advised by LIA & ERIC will do the heavy lifting for all
three tasks. He/she will have to collaborate with the Master’s interns involved in WP1. To ease this
collaboration, our schedule is built so that the internship periods overlap with the PhD.

Deliverables:

• Open access collection of networks and related statistics.
• Open-source software implementing the graph prediction benchmark.
• Open-source software implementing original graph prediction methods.